string16.h revision 3345a6884c488ff3a535c2c9acdd33d74b37e311
1c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// Copyright (c) 2006-2008 The Chromium Authors. All rights reserved. 2c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// Use of this source code is governed by a BSD-style license that can be 3c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// found in the LICENSE file. 4c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 5c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott#ifndef BASE_STRING16_H_ 6c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott#define BASE_STRING16_H_ 73345a6884c488ff3a535c2c9acdd33d74b37e311Iain Merrick#pragma once 8c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 9c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// WHAT: 10c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// A version of std::basic_string that provides 2-byte characters even when 11c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// wchar_t is not implemented as a 2-byte type. You can access this class as 12c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// string16. We also define char16, which string16 is based upon. 13c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// 14c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// WHY: 15c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// On Windows, wchar_t is 2 bytes, and it can conveniently handle UTF-16/UCS-2 16c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// data. Plenty of existing code operates on strings encoded as UTF-16. 17c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// 18c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// On many other platforms, sizeof(wchar_t) is 4 bytes by default. We can make 19c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// it 2 bytes by using the GCC flag -fshort-wchar. But then std::wstring fails 20c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// at run time, because it calls some functions (like wcslen) that come from 21c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// the system's native C library -- which was built with a 4-byte wchar_t! 22c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// It's wasteful to use 4-byte wchar_t strings to carry UTF-16 data, and it's 23c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// entirely improper on those systems where the encoding of wchar_t is defined 24c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// as UTF-32. 25c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// 26c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// Here, we define string16, which is similar to std::wstring but replaces all 27c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// libc functions with custom, 2-byte-char compatible routines. It is capable 28c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// of carrying UTF-16-encoded data. 29c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 30c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott#include <stdio.h> 31c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott#include <string> 32c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 33c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott#include "base/basictypes.h" 34c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 35c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott#if defined(WCHAR_T_IS_UTF16) 36c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 37c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scotttypedef wchar_t char16; 38c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scotttypedef std::wstring string16; 39c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 40c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott#elif defined(WCHAR_T_IS_UTF32) 41c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 42c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scotttypedef uint16 char16; 43c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 44c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scottnamespace base { 45c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 46c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// char16 versions of the functions required by string16_char_traits; these 47c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// are based on the wide character functions of similar names ("w" or "wcs" 48c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// instead of "c16"). 49c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scottint c16memcmp(const char16* s1, const char16* s2, size_t n); 50c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scottsize_t c16len(const char16* s); 51c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scottconst char16* c16memchr(const char16* s, char16 c, size_t n); 52c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scottchar16* c16memmove(char16* s1, const char16* s2, size_t n); 53c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scottchar16* c16memcpy(char16* s1, const char16* s2, size_t n); 54c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scottchar16* c16memset(char16* s, char16 c, size_t n); 55c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 56c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scottstruct string16_char_traits { 57c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott typedef char16 char_type; 58c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott typedef int int_type; 59c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 60c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott // int_type needs to be able to hold each possible value of char_type, and in 61c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott // addition, the distinct value of eof(). 62c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott COMPILE_ASSERT(sizeof(int_type) > sizeof(char_type), unexpected_type_width); 63c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 64c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott typedef std::streamoff off_type; 65c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott typedef mbstate_t state_type; 66c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott typedef std::fpos<state_type> pos_type; 67c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 68c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static void assign(char_type& c1, const char_type& c2) { 69c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott c1 = c2; 70c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 71c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 72c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static bool eq(const char_type& c1, const char_type& c2) { 73c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return c1 == c2; 74c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 75c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static bool lt(const char_type& c1, const char_type& c2) { 76c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return c1 < c2; 77c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 78c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 79c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static int compare(const char_type* s1, const char_type* s2, size_t n) { 80c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return c16memcmp(s1, s2, n); 81c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 82c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 83c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static size_t length(const char_type* s) { 84c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return c16len(s); 85c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 86c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 87c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static const char_type* find(const char_type* s, size_t n, 88c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott const char_type& a) { 89c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return c16memchr(s, a, n); 90c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 91c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 92c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static char_type* move(char_type* s1, const char_type* s2, int_type n) { 93c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return c16memmove(s1, s2, n); 94c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 95c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 96c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static char_type* copy(char_type* s1, const char_type* s2, size_t n) { 97c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return c16memcpy(s1, s2, n); 98c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 99c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 100c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static char_type* assign(char_type* s, size_t n, char_type a) { 101c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return c16memset(s, a, n); 102c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 103c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 104c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static int_type not_eof(const int_type& c) { 105c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return eq_int_type(c, eof()) ? 0 : c; 106c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 107c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 108c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static char_type to_char_type(const int_type& c) { 109c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return char_type(c); 110c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 111c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 112c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static int_type to_int_type(const char_type& c) { 113c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return int_type(c); 114c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 115c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 116c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static bool eq_int_type(const int_type& c1, const int_type& c2) { 117c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return c1 == c2; 118c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 119c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 120c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott static int_type eof() { 121c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott return static_cast<int_type>(EOF); 122c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott } 123c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott}; 124c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 125c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott} // namespace base 126c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 127c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// The string class will be explicitly instantiated only once, in string16.cc. 128c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// 129c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// std::basic_string<> in GNU libstdc++ contains a static data member, 130c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// _S_empty_rep_storage, to represent empty strings. When an operation such 131c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// as assignment or destruction is performed on a string, causing its existing 132c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// data member to be invalidated, it must not be freed if this static data 133c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// member is being used. Otherwise, it counts as an attempt to free static 134c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// (and not allocated) data, which is a memory error. 135c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// 136c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// Generally, due to C++ template magic, _S_empty_rep_storage will be marked 137c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// as a coalesced symbol, meaning that the linker will combine multiple 138c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// instances into a single one when generating output. 139c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// 140c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// If a string class is used by multiple shared libraries, a problem occurs. 141c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// Each library will get its own copy of _S_empty_rep_storage. When strings 142c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// are passed across a library boundary for alteration or destruction, memory 143c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// errors will result. GNU libstdc++ contains a configuration option, 144c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// --enable-fully-dynamic-string (_GLIBCXX_FULLY_DYNAMIC_STRING), which 145c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// disables the static data member optimization, but it's a good optimization 146c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// and non-STL code is generally at the mercy of the system's STL 147c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// configuration. Fully-dynamic strings are not the default for GNU libstdc++ 148c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// libstdc++ itself or for the libstdc++ installations on the systems we care 149c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// about, such as Mac OS X and relevant flavors of Linux. 150c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// 151c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// See also http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24196 . 152c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// 153c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// To avoid problems, string classes need to be explicitly instantiated only 154c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// once, in exactly one library. All other string users see it via an "extern" 155c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// declaration. This is precisely how GNU libstdc++ handles 156c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// std::basic_string<char> (string) and std::basic_string<wchar_t> (wstring). 157c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// 158c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// This also works around a Mac OS X linker bug in ld64-85.2.1 (Xcode 3.1.2), 159c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// in which the linker does not fully coalesce symbols when dead code 160c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// stripping is enabled. This bug causes the memory errors described above 161c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// to occur even when a std::basic_string<> does not cross shared library 162c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// boundaries, such as in statically-linked executables. 163c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// 164c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott// TODO(mark): File this bug with Apple and update this note with a bug number. 165c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 166c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scottextern template class std::basic_string<char16, base::string16_char_traits>; 167c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 168c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scotttypedef std::basic_string<char16, base::string16_char_traits> string16; 169c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 170c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scottextern std::ostream& operator<<(std::ostream& out, const string16& str); 171c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 172c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott#endif // WCHAR_T_IS_UTF32 173c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott 174c7f5f8508d98d5952d42ed7648c2a8f30a4da156Patrick Scott#endif // BASE_STRING16_H_ 175